如何从测试集的最后一个数据点进行预测

Question

我正在进行一个时间序列预测项目。我的任务是在拥有 1 月至 11 月的数据时预测 12 月的销售额。我将数据分成训练集和测试集。我已经应用了随机森林回归来预测测试集。但是，我不知道如何使用该模型来预测 12 月份的销售额。你能告诉我怎么做吗？提前谢谢你。

Answer 1

如果您已经完成了对数据的清理，并且已经将它们拆分为 training 和 testing 数据集。您可以简单地将它们放入我创建的 pipline 函数中。这个generic function 将任何算法和数据作为输入并制作模型，执行交叉验证并为testing 数据集生成预测。

from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error
import pandas as pd
import plotly.plotly as ply
import cufflinks as cf

cf.go_offline()


#Define target and ID columns:
target = 'sales'
IDcol = ['months']
predictors = [x for x in training.columns if x not in [target]+IDcol]

alg = RandomForestRegressor(n_estimators=200,max_depth=5, min_samples_leaf=100,n_jobs=4)
test = modelfitting(alg, training, testing, predictors, target)
coef5 = pd.Series(alg.feature_importances_, predictors).sort_values(ascending=False)
coef5.iplot(kind='bar', title='Feature Importances')

for_plot = test
for_plot = for_plot[['sales prediction']]
for_plot.iplot()


def modelfitting(alg, training, testing, predictors, target):
    # Fit the algorithm on the data
    alg.fit(training[predictors], training[target])

    # Predict training set:
    dtrain_predictions = alg.predict(training[predictors])

    # Perform cross-validation:
    cv_score = cross_val_score(alg, training[predictors], training[target], cv=20, scoring='neg_mean_squared_error')
    cv_score = np.sqrt(np.abs(cv_score))

    # Print model report:
    print "\nModel Report"
    print "RMSE : %.4g" % np.sqrt(metrics.mean_squared_error(training[target].values, dtrain_predictions))
    print "CV Score : Mean - %.4g | Std - %.4g | Min - %.4g | Max - %.4g" % (
    np.mean(cv_score), np.std(cv_score), np.min(cv_score), np.max(cv_score))

    # Predict on testing data:
    testing["sales prediction"] = alg.predict(testing[predictors])

    return testing

我已经放入了不言自明的 cmets。如果您在理解代码方面遇到困难，请随时在 cmets 中讨论。